Nullsignal

Autonomous scientific research, applied to quantitative trading.

Nullsignal is an autonomous research and quantitative trading platform. Its agentic researchers formulate hypotheses, design and run experiments, interpret results and decide what to investigate next. I develop both the research agents and the harnesses that give them the tools, experimental context and boundaries to work independently.

Nullsignal

My roleAgentic research systems · Quantitative research · Platform engineering

In the product

Nullsignal Agents page with navigation sidebar, research programmes and current, recent and planned experiments
The Agents page brings research programmes and current, recent and planned experiments into one view, alongside the wider operations console. View full size →

The problem

An autonomous researcher needs to do more than generate ideas or run code. It needs to pursue a question across many experiments, compare results fairly, retain what it has learned and change direction when the evidence warrants it. The challenge is to make that work scientifically useful without requiring a person to direct every step.

Trading provides a demanding test bed. A promising backtest must survive realistic costs and evaluation on data not used to develop the strategy. A research finding must then translate into a working system, where strategy decisions, orders and actual fills can differ.

My role

I develop the research agents and their execution harnesses alongside the quantitative platform: market-data processing, machine-learning risk models, backtesting and trading infrastructure. My work connects the scientific process with the software needed to carry it out, evaluate its conclusions and put reviewed improvements into use.

The solution

The researcher works through an agent harness: the surrounding software that gives it access to experiment tools, preserves research context, schedules computation and enforces budgets and permissions. Within an approved research programme, agents can propose hypotheses, design and run experiments, interpret the evidence and choose their next investigation.

The harnesses make that work traceable and evaluable. Experiment history is retained, candidates are compared with controls, and held-out data is protected for evaluation. Negative results can close a line of inquiry rather than simply trigger another attempt to find a favourable answer. This supports sustained research with human direction at programme and review points, rather than at every experimental step.

The trading platform provides the application: a Python Signal Engine produces strategy decisions, while a separate Trade Manager handles orders and position protection. After human review, research improvements can automatically be deployed to the Signal Engine in production.

An autonomous research cycle, guided by evidence
  1. Form a hypothesisThe researcher chooses a question within an approved programme and budget.
  2. Run an experimentThe harness provides tools, computation and retained experimental evidence.
  3. Interpret and continueResults guide the next investigation; reviewed improvements can move into production.

Current status

Nullsignal continues to evolve both as an autonomous research system and as a quantitative trading platform. The focus is on the quality of the research process and its engineering, rather than a claim of proven investment returns.

The same approach could support quantitative optimisation in insurance, medicine and other industries. Applying it in those settings would require suitable data, domain expertise and independent validation.

Explore further

Discuss Nullsignal

Interested in autonomous research for a different domain?